Discriminative Non-blind Deblurring

Abstract

Non-blind deblurring is an integral component of blind approaches for removing image blur due to camera shake. Even though learning-based deblurring methods exist, they have been limited to the generative case and are computationally expensive. To this date, manually-defined models are thus most widely used, though limiting the attained restoration quality. We address this gap by proposing a discriminative approach for non-blind deblurring. One key challenge is that the blur kernel in use at test time is not known in advance.To address this, we analyze existing approaches that use half-quadratic regularization. From this analysis, we derive a discriminative model cascade for image deblurring. Our cascade model consists of a Gaussian CRF at each stage, based on the recently introduced regression tree fields. We train our model by loss minimization and use synthetically generated blur kernels to generate training data. Our experiments show that the proposed approach is efficient and yields state-of-the-art restoration quality on images corrupted with synthetic and real blur.

Uncertainty-Driven Forest Predictors for Vertebra Localization and Segmentation

“…This modeling framework is however static, as it separates feature generation from inference (i.e., "model fitting"). It has been shown that better features can be generated by interleaving feature generation with MAP inference [9,11,7]. 4 In this work we take this idea a step further: Instead of interleaving feature generation with a pixel-level structured model or model-agnostic smoothing, we Shared first authors.…”

Discriminative Indexing for Probabilistic Image Patch Priors

“…We can see our nonblind deblurring component improves the performance of both Cho's and Xu's approaches in most cases. Although our PSNR is slightly worse than Schmidt's approach [11], the running time per each RGB image is 2 minites in average, which is about 20 times faster than [11] 2 and 40 times faster than EPLL. Also note that this is achieved when the blur kernel for index construction is dramatically different from the blur kernels used in the test images.…”

“…Different deblurring approaches are applied on the input images, and average PSNRs among all the kernels on each image are reported as quantitative measurements. We compare our deblurring approach with treebased indexing with several state-of-the-art algorithms, including Discriminative non-blind deblurring [11] (referred as Schmidt), 0 based deblurring [16] (referred as Xu), and Cho's fast deblurring [17] (referred as Cho).…”

“…This method can be interpreted as using random forests to pre-index a flexible prior defined on cliques in the random field. A similar idea of using pre-trained tree structures to efficiently construct a Regression Tree Field is also used in deblurring recently [11].…”

“…9 end 10 Collect all the candidate w and b, store the one maximizing Equation (15) in the tree node T . 11 Train the left and right child of T with (D − 1) tree depth and Y+, Y− as training data, which are defined in Equation (13). 12 return T.…”